app.py

from flask import Flask, request, render_template, jsonify
from huggingface_hub import hf_hub_download
import cv2
import numpy as np
import tensorflow as tf
from tensorflow import keras
import pandas as pd
import base64
import json

# Initialize Flask app
app = Flask(__name__)

# Load model and data at startup
model = keras.models.load_model("./sneaker_category_predictor_v2.keras")
# model = tf.keras.models.load_model("./sneaker_category_predictor_v2.keras")
# model_path = hf_hub_download(repo_id="thejagstudio/SneakerAI", filename="sneaker_category_predictor_v2.h5", repo_type="model")
# model = tf.keras.models.load_model(model_path)
# model = load_model(model_path)

# Define expected columns for one-hot encoding
with open("metadata.json", "r") as f:
	METADATA_COLUMNS = json.load(f)
with open("productData.json", "r") as f:
	products = json.load(f)


def encode_metadata(data):
    # Create DataFrame with single row
    df = pd.DataFrame({k: [v.lower()] for k, v in data.items()})

    # Initialize empty DataFrame with all possible columns
    encoded = pd.DataFrame()

    # Encode each feature maintaining consistent columns
    for feature, possible_values in METADATA_COLUMNS.items():
        feature_encoded = pd.get_dummies(df[feature], prefix=feature)
        # Add missing columns with 0s
        for value in possible_values:
            col_name = f"{feature}_{value}"
            if col_name not in feature_encoded.columns:
                feature_encoded[col_name] = 0
        encoded = pd.concat([encoded, feature_encoded], axis=1)

    # Ensure consistent column order
    all_columns = [
        f"{feat}_{val}" for feat, vals in METADATA_COLUMNS.items() for val in vals
    ]
    encoded = encoded.reindex(columns=all_columns, fill_value=0)

    return encoded.values.astype(np.float32)


@app.route("/")
def index():
    global METADATA_COLUMNS,products
    # 6 random products
    random_products = np.random.choice(products, 6, replace=False).tolist()
    productsString = json.dumps(random_products,indent=4)
    return render_template("index.html", metadata=METADATA_COLUMNS,products=random_products,productsString=productsString)


@app.route("/predict", methods=["POST"])
def predict():
	try:
		data = request.json

		# Process image
		img_data = base64.b64decode(data["image"])
		img_array = np.frombuffer(img_data, np.uint8)
		img = cv2.imdecode(img_array, cv2.IMREAD_COLOR)
		img = cv2.resize(img, (224, 224))
		img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
		img = img / 255.0
		img = np.expand_dims(img, axis=0)

		# Encode metadata with consistent columns
		metadata = encode_metadata(
			{
				"brand": data["brand"],
				"color": data["color"],
				"gender": data["gender"],
				"midsole": data["midsole"],
				"upperMaterial": data["upperMaterial"],
			}
		)
		# Make prediction
		predictions = model.predict([img, metadata])
		categories = ['Basketball', 'Boot', 'Cleat', 'Lifestyle', 'Other', 'Running', 'Sandal', 'Skateboarding']
		confidenceList = predictions[0].tolist()

		return jsonify({"categories": categories, "confidence": confidenceList})

	except Exception as e:
		return jsonify({"error": str(e)}), 400


if __name__ == "__main__":
    app.run(host="0.0.0.0", port=7860)